Enhanced Soil Liquefaction Potential Estimation using Machine Learning and Web-Based Platform

Abstract

In this study, a new web-based platform was developed for fast estimation of soil liquefaction potential (SLP). The geotechnical results from 47 boreholes in the north of Iran were collected over three years to create an estimator model. The dataset included information on SPT, soil type, strength parameters, and water content. Python libraries Pycaret and Gradio were used to develop the model for SLP. A set of pipeline codes were applied to base classifiers, including 13 different machine learning models such as the Ada boost classifier (ad), decision tree classifier (dt), gradient boosting classifier (gb), the k-neighbors classifier (knn), light gradient boosting machine (lightgbm) and random forest classifier (rf). The results show that the lightgbm model outperformed the other applied machine learning classifiers with accuracy = 0.946, AUC = 0.982, and F1-score = 0.9. The proposed model was then used as the primary element of the web-based application, providing a helpful tool for geotechnical engineers to determine SLP.